Streaming Codes For Variable-Size Arrivals

Abstract

Streaming codes are a class of convolutional codes that encode a stream of source packets, which arrive sequentially with a strict decoding-delay constraint, for transmission over a packet-erasure channel. A number of recent works have proposed capacity achieving streaming code constructions for adversarial packet-erasure channels that include bursts and arbitrary erasures. The existing models for streaming codes, to the best of our knowledge, consider a setting where all the source packets are of the same, fixed size. However, live video streaming applications have source packets with highly variable sizes. Motivated by this application, in this paper, we present a generalized model for streaming codes that incorporates variable-size arrivals. The variability in the size of the source packets induces a new trade-off between the rate of the code and the minimum decoding delay when the channel introduces no erasures. We also present a simple construction of streaming codes for variable-size arrivals, which builds upon existing streaming code constructions designed for fixed-size arrivals. We theoretically characterize the rate achieved by the proposed code when arrival sizes are drawn independently from any distribution with a finite support. Furthermore, using a live video trace and several representative parameter settings, we show that proposed code construction achieves a rate of approximately 89.5% of an upper bound and achieves a significantly higher rate (10%-50%) than naively using the existing streaming codes designed for fixed-size arrivals.